Predicting stock market direction in South African banking sector using ensemble machine learning techniques

Abstract

<abstract> <p>The ability to accurately predict stock price direction is important for investors and policymakers. We aim to predict the direction of daily stock returns for five major South African banks using ensemble machine learning techniques. Financial ratios were used as predictors in single classifier and ensemble models. The key findings were that the support vector machine performed best among single classifiers, with the highest accuracy for 4 banks ranging from 54% to 99% and produces fewer wrong classifications compared to its peer single classifiers. More importantly, the heterogeneous ensemble classifier, combining support vector machines, decision trees and k- (KNN) nearest neighbors, achieved average accuracy rates above 95% and outperformed all other models. This confirms that ensemble methods that combine multiple models can generate more accurate predictions compared to single classifiers. The results suggest that the heterogeneous ensemble is a suitable approach for predicting stock price direction in the South African banking sector. The findings imply that investing in banks may be a good decision and can assist investors. However, further research could expand the models to incorporate macroeconomic and other external factors that influence stock prices. Overall, we demonstrate the value of ensemble learning for a complex forecasting problem. The heterogeneous ensemble approach achieved high accuracy and outperformed single classifiers. However, future research incorporating additional factors and policy implications could build on these findings.</p> </abstract>